Technology-Assisted Review (TAR)

Technology-Assisted Review (TAR) is a workflow in which reviewers code a sample of documents as relevant or not, and an algorithm learns from those judgments to predict how the remaining population should be classified. Often called predictive coding, it lets organizations sort millions of records far faster than purely manual, document-by-document review. In recordkeeping and e-discovery, TAR matters because the cost and timeline of reviewing electronically stored information can otherwise be prohibitive, and defensible, well-documented TAR processes have been accepted by courts as proportional and reasonable. A common distinction is between “TAR 1.0,” which trains on a fixed seed set before review begins, and “TAR 2.0,” or continuous active learning, where the model keeps retraining as reviewers work. For example, a litigation-hold collection of half a million emails might be ranked so that the most likely responsive items surface first, cutting the volume that humans must examine. To remain defensible, teams should preserve provenance and metadata about training decisions, validation sampling, and recall estimates, since the methodology itself may be challenged.